Bondi w/vertical B sometimes fails in CI

[bprather]

Our Bondi problem seeded with a unifom vertical field sometimes initializes a field with a divergence. This appears to be inconsistent from run to run.

Since the field is the only one among the default parameter files to stretch into the ghost zones, I'm assuming this is a race condition or weird potentially uninitialized memory in how we handle ghost zones during init and/or freeze in Dirichlet boundaries using the existing ghost values (potentially related to #76). This might be relevant for multi-zone runs, though most initialization in those is more careful than the throwaway test here, and the problem is very easily visible if it happens.

For now, workaround is to re-run KHARMA or the relevant CI test.

[bprather]

This was probably fixed when I updated the B field init in #96. Indeed I haven't seen it since in any CI runs, so, closing